The latest marketing tactic on LinkedIn: AI-generated faces Stanford researchers uncovered more than 1,000 of these LinkedIn profiles. A technology that has been used to promote misinformation online has now entered the corporate world.

That smiling LinkedIn profile face might be a computer-generated fake

  • Download
  • <iframe src="https://www.npr.org/player/embed/1088140809/1089885326" width="100%" height="290" frameborder="0" scrolling="no" title="NPR embedded audio player">
  • Transcript

A MARTINEZ, HOST:

As long as there have been salespeople, there have been shady sales tactics, from the hard sell on a used car lot to those robocalls warning you about that car's warranty. Now they come with a 21st century twist - fake social media accounts using computer-generated photos. The professional networking site LinkedIn has taken down hundreds of these fake profiles following an investigation by Stanford University and NPR. Here's NPR tech correspondent Shannon Bond.

SHANNON BOND, BYLINE: Renee DiResta usually ignores LinkedIn messages like this one. The sender, Keenan Ramsey, mentioned they both belong to a LinkedIn group for entrepreneurs before pivoting to a pitch for software. But something about Ramsey's profile picture made DiResta pause.

RENEE DIRESTA: The face jumped out at me as being fake.

BOND: It looked like a typical corporate headshot, but little details were off.

DIRESTA: She has mismatched earrings. Now, I myself have mismatched earrings (laughter), and I'm real. So that's not - you know, that's not the be all, end all. But you can see the way her hair sort of blurs into the background. There's pieces that just kind of come out of nowhere. On one side, it looks almost like it breaks off, but then comes back at the bottom.

BOND: Also, Ramsey's eyes were aligned perfectly in the middle of the image. Individually, these things could all be explained away. But together, they made DiResta suspicious. She happens to be a researcher at the Stanford Internet Observatory who's studied Russian disinformation campaigns and anti-vaccine conspiracies. And to her trained eye, these anomalies were red flags that the photo had likely been created by artificial intelligence.

DIRESTA: You know, in the course of my work, I look at a lot of these things, mostly in the context of political influence operations. But all of a sudden, here was a fake person in my inbox.

BOND: That message launched DiResta and her colleague Josh Goldstein on an investigation that uncovered more than 1,000 LinkedIn profiles with these apparently fake faces. Similar fake social media accounts have been used to push Russian propaganda about Ukraine and spread Chinese disinformation on Facebook and Twitter. But these profiles on LinkedIn - they're being used to spam people about products and services. Here's how it works. The profiles claim to work for more than 70 different companies. They send LinkedIn messages to potential customers, like the one DiResta got from Ramsey, pitching software from a company called RingCentral.

DIRESTA: Their message was inviting me to set up a time to talk about RingCentral for my startup, which I no longer have.

BOND: Anyone who takes the bait gets connected to a real person who tries to close the deal. Now, NPR didn't find anything illegal, and in some ways, it's not surprising that people are now using the latest technology to get an edge on LinkedIn, which is all about professional networking. The pandemic made it harder for many salespeople to do their jobs in person, so there's a lot of demand to make virtual connections. And creating profiles with fake faces is tempting. It's cheaper than hiring real people. You can't immediately debunk them with a reverse image search, and the faces are really convincing.

HANY FARID: They are now as realistic as real faces and slightly more trustworthy, which is super interesting and really creepy.

BOND: That's Hany Farid, a digital forensic expert at the University of California, Berkeley. He co-authored a recent study that found people have just a 50% chance of guessing correctly whether a face was created by a computer. That's no better than flipping a coin. He also found that people say they trust faces made by AI a bit more than real ones. Farid thinks that's because the software tends to stick to the most average features when generating a face.

FARI: It turns out that the average face - if you take a bunch of faces and align them and compute an average of all of them, that face tends to look trustworthy. Because it's familiar, right? It's - looks like something, like, of somebody we know.

BOND: These days, fake faces are widely available. You can go online right now and download them for free. And it's this proliferation of AI-generated content that worries Farid. Because while using synthetic salespeople to send pitches on LinkedIn might not seem all that serious...

FARI: It is the canary in the coal mine.

BOND: And the way this technology is advancing, today's trustworthy fake face could quickly become tomorrow's fraudulent video.

MARTINEZ: And Shannon is here to tell us more about these fake accounts, starting with - actually, starting with - these fake faces, Shannon, are terrifying me. It's like a Jordan Peele movie, to tell you the truth. But who's behind all these things?

BOND: Yeah, A, that was my first question, too. So I called up RingCentral, the company where dozens of these profiles claim to work, including that Keenan Ramsey account who first messaged Renee DiResta. And a spokesperson said RingCentral had never heard of any of them. Another previous employer listed on Ramsey's LinkedIn profile said the same thing. And NYU, where she says she got an undergraduate business degree, doesn't have any record of her either. So what RingCentral told me is, it's hired other companies to reach out to potential customers about its software. And RingCentral says one of them created the fake profiles. Here's what Heather Hinton, RingCentral's chief information security officer, told me.

HEATHER HINTON: There was something going on that we hadn't been aware of. And so we've gone and - you know, this is not how we do business.

BOND: And this is what I heard from many of the companies these profiles claimed to work for. They had hired outside marketers to help with sales. Now, that's not unusual or suspicious, but these companies said they had not approved any use of computer-generated images.

MARTINEZ: Yeah, outside marketers, outside companies - who are they?

BOND: Yeah, these are people who sell LinkedIn marketing services. And I tried contacting a lot of them. Most would not talk to me. One I did hear from was called airSales. It told me it works with freelancers who might make LinkedIn profiles, but that airSales was not involved. We also found the website of another one, called LIA, advertising what it calls AI-generated LinkedIn avatars for $300 a month. They say they have hundreds of these avatars, which look an awful lot like the fake accounts Stanford found. And these are just ready for a client to choose from. Now, no one in LIA responded at my attempts to contact them, and it's now scrubbed all information from its website. And ultimately, we weren't able to verify who made these profiles or find anyone who authorized them.

MARTINEZ: What does LinkedIn have to say about this?

BOND: Well, it's taken down most of the accounts the Stanford team found, and it says it has rules against creating fake accounts or falsifying information. People are supposed to use real pictures of themselves. And LinkedIn says it's constantly improving its detection models to catch profiles with fake faces.

MARTINEZ: Shannon, do we know how common these fake profiles are on LinkedIn?

BOND: Well, LinkedIn has 810 million members. And, you know, Stanford identified around a thousand fake accounts, so that's a small portion. But, A, since my story published, I've heard from lots of people saying they've encountered profiles similar to these being used for other purposes on LinkedIn, like recruiting them for jobs. So I think it's likely there are a lot more of these on the platform than we've identified here.

MARTINEZ: So just how suspicious should we all be about people reaching out on LinkedIn?

BOND: Well, you know, all social media platforms have problems with fake accounts. What's different here, though, is these AI-generated faces. You know, they're just - they're so realistic and they're just so available. It's easy for anyone to use them. So it's good advice in general to be skeptical about what you're seeing online. You know, if someone's reaching out to you - asking who they are and what their motivation for contacting you is.

MARTINEZ: I'm going to stare deep into those AI-generated eyes. That's NPR tech correspondent Shannon Bond. Shannon, thanks.

BOND: Thanks, A.

MARTINEZ: And just a note - LinkedIn and its parent company, Microsoft, are among NPR's financial supporters.

(SOUNDBITE OF SMARTFACE'S "TOO TIRED")

Copyright © 2022 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.

NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.